What Is an LLM Referrer?

An LLM referrer is the Referer HTTP header (or the parsed source in your analytics) that identifies an AI platform as the origin of a click. The common strings are chatgpt.com and openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, and claude.ai. Capturing these cleanly is the foundation of measuring AI referral traffic, but the header is inconsistently sent, so no capture method is complete.

Which referrer strings should you watch?

Gemini and Copilot blur into their parent search hosts, so Google-referred AI clicks can be indistinguishable from classic organic without landing-page or query context.

How do you capture LLM referrers cleanly?

• Log the raw Referer header at the server or edge, before client-side analytics can drop it.
• Maintain an allowlist of AI hostnames and normalize subdomains to a single label per platform.
• Cross-check against Google Search Console and Bing Webmaster data for AI-surface impressions the referrer misses.
• Treat referrer-less-but-plausible sessions (direct visits to deep pages spiking after a citation) as probable AI traffic.

Why is it never complete?

Referrer-policy headers, in-app browsers, and privacy modes blank the string on a meaningful share of clicks. The GEO reality is that a chunk of AI-driven traffic always lands as dark traffic, which is why self-reported attribution surveys remain the practical catch-all.

Example

An edge worker logs every inbound Referer, tags any match to perplexity.ai, and writes it to a dedicated table. Over a quarter it captures 60% of known Perplexity clicks; the remaining 40% show as direct — recovered only when the sign-up form asks how the user found the product.